An important performance characteristic of imaging devices is dynamic range. A large dynamic range is desirable in applications for sensing low light signals and capturing images with large variations in illuminance or brightness.
In particular, the dynamic range of an image sensor can be defined as the ratio of the minimum illuminance the sensor detects under saturation to the illuminance the sensor detects at signal-to-noise ratio (SNR) equal to 1. The dynamic range of a scene can also be expressed as the ratio of its highest illumination level to its lowest illumination level.
Intrascene dynamic range refers to the range of incident signal that can be accommodated by a sensor in a single frame of imager data. Examples of scenes that generate high dynamic range incident signals include an indoor room with outdoor window, outdoor mixed shadow and bright sunshine, night time scenes combining artificial lighting and shadows, and in automotive context, an auto entering or about to leave a tunnel or shadowed area on a bright day.
Many different types of approaches for creating devices with high dynamic range have been described in the literature. A common denominator of most approaches rely on performing companding within the pixel by a having either a total conversion to a log scale (so-called logarithmic pixel) or a mixed linear and logarithmic response region in the pixel. These approaches have several major drawbacks. First, the knee point in linear-to-log transition is difficult to control leading to fixed pattern noise in the output image. Second, under low light the log portion of the circuit is slow to respond leading to lag. Third, a logarithmic representation of the signal in the voltage domain (or charge domain) means that small variations in signal due to fixed pattern noise leads to large variations in represented signal.
Linear approaches have also been used to increase dynamic range where the integration time is varied during a frame capture to generate several different integrated pixel signals. In the context of a CMOS pixel, integration time refers to the time period during which a capacitor or charge well accumulates a charge or discharges a voltage from a pre-charge level (from a reset voltage) as a result of exposure of a photosensor to incident light. The integrated signal is then read-out and sampled. If a CMOS pixel's stored charge rises or falls to a point where it cannot further increase or decrease during the integration period, then it is said that the CMOS pixel has reached its saturation point. Conventional implementations which vary integration time during frame capture have require additional logic and memory structures to store data generated by reading out the pixel at different points in time and thus are less than optimal as a design choice.
FIG. 1 shows how changes in integration time affects the magnitude of light intensity which a CMOS sensor can absorb without reaching the saturation voltage 1 thereby avoiding loss of image data. In particular, the FIG. 1 example demonstrates the behavior of the output signal from a pixel with a long integration time 2 and a short integration time 3.
Capturing still images with different integration times and then merging them is an effective way to extend the dynamic range of a linear sensor without losing contrast at high light level, in a manner similar to how nonlinear sensors perform. For a linear sensor, a signal output S is proportional to light intensity and integration time. With a constant light input I over an integration time Tint, the signal output can be expressed asS=ks·I·Tint  (1)                where ks is the pixel's sensitivity.        
For the example shown in FIG. 1, with one integration period 4, the sensor's dynamic range is independent of integration time, which is
                    DR        =                                            Saturation              ⁢                                                          ⁢              voltage                                      Read              ⁢                                                          ⁢              noise                                =                                                    I                                  L                  ⁢                                                                          ⁢                  max                                                            I                                  L                  ⁢                                                                          ⁢                  min                                                      =                                          I                                  S                  ⁢                                                                          ⁢                  max                                                            I                                  S                  ⁢                                                                          ⁢                  min                                                                                        (        2        )                            where ILmax 5 is the minimum light intensity which causes the pixel to saturate with integration time TL. ILmin 6 is the light intensity when signal output equals read noise with integration time TL. ISmax 7 is the minimum light intensity which causes the pixel to saturate with integration time TS. ISmin 8 is the light intensity when signal output equals read noise with integration time TS. With two integration times (i.e., range 9), the extended dynamic range DRext can be expressed as        
                              DR          ext                =                                            I                              S                ⁢                                                                  ⁢                max                                                    I                              L                ⁢                                                                  ⁢                min                                              =                                                    T                L                                            T                S                                      ·            DR                                              (        3        )            
Accordingly, dynamic range (DR) is extended by the ratio of the long integration time to the short integration time. For example, if long integration time is 20 and short integration time is 4, then DR is be multiplied by a factor of 5.
A multiple integration approach was first used in CCD sensors to increase dynamic range. A similar approach was used in CMOS active pixel sensors and in charge multiplication devices (CMD) and since its initial use, the multiple integration approach has become one of the most commonly used techniques in high dynamic range sensors.
A conventional high dynamic range imager uses two sample and hold circuits: one is a linear sample and hold circuit for each column of the array and captures a linear signal related to a difference between the pixel image output signal and a reset output signal to which the pixel is reset at the beginning of the integration period. The other is an extended dynamic range (XDR) sample and hold circuit for each column of the array which captures an XDR signal related to a difference between the pixel image output signal and an XDR reset level to which the pixel is reset at a predetermined time before the end of the integration period.
A high intrascene dynamic range CMOS active pixel sensor using dual sampling has been previously created but has a number of shortcomings. For example, a second column signal processing chain circuit and associated sample and hold circuit must be added to the upper part of the CMOS sensor. During operation, row n is first selected for read out and copied into a lower sample and hold circuit. Row n is reset in the process. Immediately after row n is read out, row n−Δ is then selected and sampled into the upper sample and hold circuit. Row n−Δ is also reset as a consequence of being copied. Both sample and hold circuits are then scanned to read out stored data. After the two sample and hold circuits are read out, the row address increases by one, and the whole process starts over again. In this readout scheme, the second readout always lags Δ rows behind the first read out. If integration time is defined for the pixels copied to the lower sample and hold circuit as T1int, while the integration time for pixels copied to the upper sample and hold circuit as T2int, the ratio of T1int T2int is (N−Δ):Δ. The intrascene dynamic range capability of the sensor is extended by the factor T1int/T2int.
There are several advantages of the dual sampling approach. First, linearity of the signal is preserved. Second, no modification to the standard CMOS APS pixel is required to achieve high dynamic range so that fill factor and pixel size can be optimized. Third, the low read noise of the CMOS APS pixel is preserved. Fourth, the extended dynamic range operation can be optionally employed, depending on control signals to the chip, without sacrificing sensor performance.
A major disadvantage of the dual sampling approach is that outputting the signal for two integration periods requires an additional analog memory on chip to synchronize these outputs. Another shortcoming is that dual sampling has not been optimally implemented for use with Phase Alternating Line (PAL) and National Television Standards Committee (NTSC) standard compliant image sensors.
The NTSC standard is the one most commonly used for video standards in North America and Japan. Europe uses PAL and the French use SECAM video standards. Both PAL and NTSC are 4:3 horizontal-to-vertical picture aspect ratios. Most television video transmitters and receivers use interlaced scanning rather than the non-interlaced progressive scanning.
Conventional dual sample image sensors using the NTSC and PAL format produce interlaced output, not progressive scan-output (non-interlaced). As shown in FIG. 2, an image frame 15 containing rows and columns of pixels is divided into two fields: an odd field (Field 1) 13 consisting of all the odd numbered rows of pixels, and the even field (Field 2) 14 consisting of all the even numbered rows of pixels. The two fields per frame scheme is known as a 2:1 interlace. Half of the frame is recorded by the odd field at time T1, and the other half of the frame is recorded by the even field at time T2. Progressive scan sensors read out a complete frame with no interlacing one row at a time. Progressive scan methods have desirable attributes such as better image capture for subjects which are moving. Thus, it would be desirable to have an increased dynamic image CMOS image sensor which is NTSC and PAL compliant and which provides a progressive scan output.